Server 2025 - high CPU and RAM

[Gostev]

Maybe consider switching to VSA then. It uses proven XFS-based storage stack with zero issues in Support in the past 5 years. XFS is simply far more mature, and with Veeam controlling the entire appliance there will never be surprises from 3rd party software either.

[jackal2001]

Installed the Dec 2025 CU but also have the following REG Keys still set.
RefsEnableLargeWorkingSetTrim -PropertyType DWord -Value 1
RefsNumberOfChunksToTrim -PropertyType DWord -Value 16

Also tried excluding MS Defender from scanning my repo volumes, disabled disk defrag that is built into windows as a scheduled task, etc. I don't even know if those reg values apply without even applying the "patch" MS provided to a select few a couple months ago.

Update, my Repo server has been up for 25 days now running since installing the patch. I have a 50TB and 30TB REFS volume using 32GB of RAM on the VM, which at this point seems to not go higher than 40% utilization.

[cbc-tgschultz]

Also updating: Things went smoothly since last time (reg key changes, official server patch via WU) except for Friday morning, when the system was stuck at 100% CPU and unusable for long enough that I just bounced it to get going again.

At this point we're already building a new set of backup storage and replication targets anyway, so we've decided to to just leverage them as Linux repositories with XFS, which seems to be not only rock solid but also much faster at block cloning anyway. Looking forward to the day that Veeam gets ZFS reflink support (outside of the TrueNAS enterprise support for it via SMB) so I can remove an abstraction layer.

[SpikeAUS]

Still running into this issue, server 2025 latest build and reg keys added.

[Didi7]

The never ending story. People have the choice.

[s.middlebrook]

Server 2022 was my choice

[AndresP]

Hi. Joined the club.... .
Server 2025 with 2026-01 CU patches ( lets say, fully patched). Proxy and REPO on RefS. No registry,etc tunings yet.
We had NO issues (with basically same setup) in other locations, and now its hitting in some locations. Looks like the crucial element there is VM (the backup object) sisze. As all works fine on smaller VM' ( this is why we hadnt experienced it before).
The Veeam Proxy/repository server just hungs ( with 100% CPU load ) on fast clone, same time memory usage is pretty normal(even the VM has more like less memory). VMTools stop responding and all went down ...
"Creating synthetic full backup (46% done) [fast clone]"

[DarrylvdPeijl]

Hi All,

Dropping in with a quick update from the Microsoft product group regarding the ReFS resource consumption on WS2025. I've been discussing this with them offline and got the go-ahead to share the latest status.

Microsoft believes they have identified the issue. They have an actual code fix in the pipeline targeted for the March servicing update. Please keep in mind the standard caveat: this target is still pending successful validation on their end.

I asked them specifically about how this fix interacts with the existing ReFS tuning parameters. Here is their official guidance:

The upcoming patch is a distinct code change, not just a tweak to the existing parameters.
If you have already applied the registry settings from the Microsoft Learn troubleshooting article for heavy ReFS memory usage, and you are happy with your current performance, leave them as is. There is no need to unset them.
Enabling those registry keys carries its own performance impact, so whatever combination is currently keeping your environment stable is what you should stick with.

Once Microsoft confirms the fix has officially made it into the March servicing package, the recommendation is to get it installed. I'll keep pushing for info and will update this thread if the timeline shifts.

Let me know if you have any specific questions so I can relay if needed.

[karsten123]

is it fixed before GTA 6 release?

[_tcpip_]

Any news on this? There's nothing about refs in the March patch notes.

This is really a slow-moving issue.....

[DarrylvdPeijl]

Just got the confirmation;

[*]The March update contains several performance and quality fixes for ReFS that should address the issues
[*](targeted) June update contains a fix for a slow memory leak

There is a Known Issue Rollback (KIR) available for the slow memory leak.
This can be installed on production environments, but it's recommended to uninstall prior to updating to the June patch level.
If you really want to test the KIR, send me an DM.

Please do share your experience!
If you still have issues, please fill out this form: https://forms.microsoft.com/e/XaY3LB2L43

[hansworst24]

Hi,

we faced similar issues like described above since July 2025. We upgraded our memory on the repository servers from 64GB to 256GB, and the problem looked solved. Although we see memory usage spike suddenly, the server recovered OK... Until recently, now we have both repository servers face the same issue of filling up all system memory again....

I couldn't find the KB March patch from Microsoft that was mentioned here above. Does someone now the KB number ? We'll then be able to validate if the issue is resolved.